v1.2 is a major SAT-focused release. We rebuilt SAT prep around original SAT-style skill practice, upgraded personalization from static routing into a full adaptive intelligence loop, and improved text-to-speech clarity so complex math is easier to follow even when your eyes are off screen.
SAT-style practice meets adaptive intelligence.
Lessons and quizzes now anchor to public SAT skill domains with original practice, while a new closed-loop personalization system tracks outcomes and actively improves future instruction.
Original SAT-style skill engine
SAT prep now relies on section and topic skill blueprints as a first-class input for both lessons and quizzes. Instead of storing official items, the system generates original passages, numbers, data, diagrams, choices, and explanations that match the requested SAT-style skill and difficulty.
Skill blueprints in-generation
SAT lesson and quiz generation now uses topic-matched SAT-style skill blueprints for exam-like structure and tone.
Original practice, no question bank
Fresh passages, numbers, tables, diagrams, answer choices, and explanations are generated without storing official source questions.
Authentic, but still personalized
Question style stays exam-true while difficulty, pacing, and explanation framing still adapt to each learner.
Adaptive personalization: closed loop
Personalization is no longer a one-time selection. v1.2 upgrades the stack to end-to-end adaptive behavior with consistent bundle and policy injection across major generation paths, anticipatory error mapping, calibrated hidden-signal thresholds, and adaptive signal weighting that learns from what actually works.
- Decision and correlation observability now follows generation into attempts, feedback, and sessions.
- Outcome-aware weighting adjusts future instruction based on real performance, not static assumptions.
- Signals are resolved consistently so competing traits do not produce noisy or contradictory prompts.
Text-to-speech now sounds substantially clearer
TTS translation and speech formatting were reworked for better clarity, smoother pacing, and stronger math readability. Complex expressions are translated into natural spoken language more reliably, so lessons are easier to understand during hands-free study.
Better math narration
Fractions, roots, symbols, and equations are normalized into more legible spoken phrasing.
Cleaner translation pipeline
Markdown and formatting artifacts are stripped more consistently before synthesis.
Pricing, reliability, and scale readiness
This release also includes pricing and model-cost tuning that increases practical usage headroom, with some workflows seeing up to 25% more usage. Alongside that, we shipped broad reliability fixes and laid infrastructure groundwork for significantly larger traffic and personalization throughput.
Usage efficiency improvements
Model and cost-path adjustments increase usable generation capacity for many learners.
Scale-focused architecture
New observability tables, indices, and policy controls prepare the stack for major growth.
Plus a broad quality pass
Across SAT prep, TTS, and personalization routes, we fixed edge cases, tightened behavior, and improved stability for day-to-day use.